[BugFix] Fix is_supported of cutlass FP8 linear (selected and fails on A100) (#55884)

Signed-off-by: Daniel Serebrenik <daserebrenik@nvidia.com>

vllm/model_executor/kernels/linear/scaled_mm/cutlass.py

@@ -17,6 +17,7 @@ from vllm.model_executor.layers.quantization.utils.quant_utils import (
 from vllm.model_executor.layers.quantization.utils.w8a8_utils import (
     CUTLASS_BLOCK_FP8_SUPPORTED,
     convert_to_channelwise,
+    cutlass_fp8_supported,
 )
 from vllm.model_executor.utils import set_weight_attrs
 from vllm.platforms import current_platform
@@ -164,8 +165,8 @@ class CutlassFP8ScaledMMLinearKernel(FP8ScaledMMLinearKernel):
     def is_supported(
         cls, compute_capability: int | None = None
     ) -> tuple[bool, str | None]:
-        if not current_platform.is_cuda():
-            return False, "requires CUDA."
+        if not cutlass_fp8_supported():
+            return False, "CUTLASS FP8 kernels not available"
         return True, None
 
     @classmethod